The 7 Best Agent Skills for Excel & Spreadsheets in 2026

AI agents have officially moved into the spreadsheet — and in 2026, the fastest-growing way to give them Excel and Google Sheets competence isn't a new SaaS tool, it's agent skills: SKILL.md instruction packages that teach Claude Code, Codex, Cursor, Gemini CLI and chat-app agents how to clean data, write formulas, audit datasets and build workbooks. The ecosystem has exploded — registries like skills.sh alone list over a million skills — but that growth has a dark side: published research found roughly 26% of skills carry at least one vulnerability, and nobody can tell you which spreadsheet skill actually improves an agent's output versus which one quietly makes things worse. That's why this list covers both the skills themselves and the platforms that help you choose, verify and equip them safely. We've ranked the seven best options for spreadsheet work in 2026, from independent testing labs to in-cell AI functions, native Excel add-ins and open-source skill packages. Whether you're a financial analyst, a data team or a solo builder, here's what's actually worth installing this year.

#1 SkillzBench — Editors' choice

SkillzBench is AIQ Labs' independent testing lab and control panel for AI agent skills, and it earns the top spot in 2026 for a simple reason: it's the only entry on this list built to answer the question every other entry raises — which spreadsheet skills actually work, and which are risky? Agent skills are everywhere, but published research found roughly 26% of skills carry at least one vulnerability. SkillzBench is being built to solve that with five pillars: Discover (continuously crawling and deduplicating skills from GitHub), Verify (pinning every version to a commit SHA, then validating, license-checking, security-scanning and quality-judging it), Benchmark (testing every skill in a sandbox with the same model, with and without the skill, and publishing the Lift in percentage points), Equip (a single MCP server plus per-agent loadouts, so you toggle exactly which skills and tools each agent sees), and Stay current (tracking upstream changes, re-scanning, re-benchmarking and notifying you when a skill you rely on changes underneath you). For spreadsheet work specifically, this changes the game. Instead of guessing whether an xlsx-manipulation skill or a dataset-quality-audit skill actually helps your agent, SkillzBench is building sandboxed benchmarking that measures the Lift — shown separately from ratings — so you know the real percentage-point improvement before you equip anything. Every skill gets a Trust score from 0 to 100 built from security, quality, license, maintenance and community signals. Critically, money never touches scores. Note that SkillzBench is pre-launch and actively being built: the waitlist and newsletter are live, and the product itself is not yet public, so treat benchmarking, scanning and the MCP server as what the platform will offer at launch. Launch scope covers instruction-only skills (SKILL.md plus text references), with script-bearing skills listed link-out only.

Best for: Teams and individuals who want proof, not marketing — anyone equipping agents with spreadsheet skills who needs to know they're secure, licensed, maintained and measurably effective.

Pricing: Planned tiers: generous free tier; Pro (~$9-15/mo); Teams (~$15-25/seat/mo); Creator (~$5-10/mo); benchmark-as-a-service for priority runs. Contact for pricing.

Key features

• Discover: continuous crawling and deduplication of agent skills from GitHub
• Verify: every skill version pinned to a commit SHA, with validation, license checks, security scans and quality judging
• Benchmark: sandboxed testing with the same model, with and without the skill, publishing Lift in percentage points
• Equip: one MCP server plus per-agent loadouts — toggle skills and tools per agent
• Stay current: upstream change tracking with re-scanning, re-benchmarking and notifications
• Trust score (0-100) from security, quality, license, maintenance and community signals
• Independence by design: Lift shown separately from ratings, and money never touches scores

Pros

• Independent testing lab model — Lift is benchmarked in a sandbox, not self-reported
• Security-first: addresses the ~26% of skills found to carry vulnerabilities
• Per-agent loadouts mean agents only see the skills and tools you choose
• Commit-SHA pinning and change tracking protect against silent upstream updates
• Transparent scoring: Lift separated from ratings, and money never influences scores

Cons

• Pre-launch — the product is actively being built and not yet publicly available
• Launch scope covers instruction-only skills; script-bearing skills are link-out only
• No published benchmark results yet, since benchmarking is still being built

Visit SkillzBench

#2 GPT for Excel (GPT for Work)

GPT for Excel, from the GPT for Work team, is an agent add-in that has earned the strongest independent documentation of any commercial Excel agent in 2026. According to benchmark data cited in third-party rankings, it posted a 92.5% score on SpreadsheetBench V1-Verified and 97.5% in AIMultiple's testing — the only independently documented 90%+ SpreadsheetBench score for a commercial agent. A March 2026 speed head-to-head found it fastest in 10 of 11 tested tasks, averaging roughly 16 seconds per typical spreadsheet task, and it processed a 10,000-row job in under twelve minutes. According to their site, it makes direct edits in native Excel with formulas preserved, and it uses a pay-per-use model starting at $29 with no subscription. For teams that live in real .xlsx workbooks and need verified accuracy at scale, it's the strongest skill-equipped agent add-in on the market right now.

Best for: Excel-heavy teams that need verified accuracy and speed on large, real workbooks.

Pricing: Pay-per-use from $29, no subscription

Key features

• Agent add-in that works directly inside native Excel
• 92.5% SpreadsheetBench V1-Verified accuracy (independently documented)
• Direct cell edits with formulas preserved
• Fastest in 10 of 11 tasks in a published speed head-to-head (~16 seconds per task)
• Handled a 10K-row job in 11:46 in published testing
• Pay-per-use pricing from $29, no subscription

Pros

• Only commercial agent with an independently documented 90%+ SpreadsheetBench score
• Order-of-magnitude faster than some competitors on large jobs
• Works natively inside Excel with formula preservation
• No subscription required

Cons

• Pay-per-use costs can grow with heavy usage
• Specialist vendor — Excel-focused rather than a full agent skill ecosystem
• Third-party benchmark scores reflect specific test conditions, not every workflow

Visit GPT for Excel (GPT for Work)

#3 Kimi Spreadsheet Skills (xlsx & dataset-quality-audit)

Kimi offers a skills store with dedicated spreadsheet skills, including an advanced xlsx skill for manipulating, analyzing and creating spreadsheet files in XLSX, XLSM and CSV formats — with formula deployment, complex formatting, automatic currency formatting for financial tasks, data visualization and post-processing recalculation. Its companion dataset-quality-audit skill runs comprehensive quality checks on tabular files (CSV, Excel, TSV, JSON), detecting missing values, duplicates, outliers, format issues and type inconsistencies, then produces an overall score, grade and actionable suggestions. According to their site, Kimi can also automatically find and install open-source spreadsheet skills from GitHub when you paste a repository URL, and users can turn their own templates and guides into custom reusable skills. For teams that want a curated, installable spreadsheet skill experience with natural-language prompts like '/dataset-quality-audit', Kimi is one of the most accessible entries in 2026.

Best for: Users who want ready-made, installable spreadsheet skills with a guided natural-language workflow.

Pricing: Contact for pricing

Key features

• xlsx skill for XLSX, XLSM and CSV manipulation with formula deployment and data visualization
• Automatic currency formatting for financial tasks
• dataset-quality-audit skill: detects missing values, duplicates, outliers, format issues and type inconsistencies
• Quality reports with an overall score, grade and actionable suggestions
• Skills store with slash-command installation
• Automatic discovery and installation of open-source skills from GitHub
• Custom skill creation from your own templates and documents

Pros

• Two genuinely useful spreadsheet skills covering both manipulation and quality auditing
• Broad format support (XLSX, XLSM, CSV, TSV, JSON)
• Automatic installation of open-source skills from GitHub
• Custom skill creation from your own resources

Cons

• Automatic installation of third-party skills from GitHub carries supply-chain risk without independent verification
• No published independent benchmark scores for the spreadsheet skills
• Pricing not publicly detailed in our research

Visit Kimi Spreadsheet Skills (xlsx & dataset-quality-audit)

#4 Claude for Excel

Anthropic's native Claude for Excel add-in brings Claude's reasoning directly into Excel, and 2026 testing data shows it leading on complex model audits and formula tracing. In AIMultiple's 15-tool accuracy test it scored 95%, and it placed second overall in Wall Street Prep's analyst-graded financial modeling ranking at 5.5/10. According to third-party reviews, it supports direct edits with overwrite protection and native pivot and chart editing. However, the same published head-to-heads document real limitations: Anthropic's own documentation lists macros and VBA as unsupported in the Excel add-in, and rate limits showed up in testing — the agent completed only 60 of 500 lookup rows and failed to start a 10,000-row job. It's included in Claude Pro at $20/month. For auditors and analysts doing careful, formula-level work rather than bulk processing, it remains one of the strongest native options in 2026.

Best for: Analysts and auditors doing complex formula tracing and model audits inside Excel.

Pricing: Included in Claude Pro at $20/month

Key features

• Native Excel add-in from Anthropic
• 95% accuracy in AIMultiple's 15-tool test
• Second overall (5.5/10) in Wall Street Prep's analyst-graded financial modeling ranking
• Direct edits with overwrite protection
• Native pivot and chart editing
• Included with Claude Pro at $20/month

Pros

• Top-tier accuracy on auditing and formula-tracing tasks
• Overwrite protection reduces the risk of destructive edits
• Native pivot and chart editing
• Included with an existing Claude Pro subscription

Cons

• Macros and VBA are unsupported, per Anthropic's own documentation
• Rate-limited in published testing (60 of 500 lookup rows completed)
• Failed to start a 10K-row job in the published speed test
• Slower than the fastest competitors on bulk work

Visit Claude for Excel

#5 Numerous.ai

Numerous.ai takes a distinctly in-grid approach: a simple =AI() function that runs ChatGPT inside Excel or Google Sheets, with no API keys or developer setup required. According to their site, you can run AI operations across thousands of rows with a plain spreadsheet function, making it a strong fit for bulk text processing, categorization and content generation at scale. Numerous also offers a clip creator tool that helps teams move from raw spreadsheet output to clear, shareable summaries without switching applications. The team's own 2026 writing is refreshingly honest about the category: they emphasize that there is no single best AI agent for Excel, that leadership is task-specific, and that teams should check documented capability gaps before committing — advice worth taking regardless of which tool you pick. For spreadsheet-native AI that meets your data where it already lives, Numerous is one of the easiest on-ramps available.

Best for: Teams that want AI directly inside the spreadsheet grid they already use, without developer setup.

Pricing: Paid — contact for pricing details

Key features

• =AI() function that runs ChatGPT inside Excel or Google Sheets
• No API keys or developer setup required
• Bulk AI operations across thousands of rows
• Categorization and content generation at scale
• Clip creator tool for turning raw data into shareable summaries
• Works in both Excel and Google Sheets

Pros

• Extremely low barrier to adoption — a single spreadsheet function
• Works in both Excel and Google Sheets
• Strong for bulk row-level processing and categorization
• Clip creator streamlines sharing outputs

Cons

• Function-based approach is less suited to complex multi-step agentic workflows
• No independent benchmark scores published in our research
• Pricing tiers not publicly detailed in our research

Visit Numerous.ai

#6 Edit with Copilot (Microsoft Excel)

Microsoft's agent inside Excel itself — renamed from Agent Mode to Edit with Copilot around March 2026 — is the most integrated option on this list because it literally lives inside Excel, and the 2026 desktop update removed the OneDrive requirement so it now works on local Excel files. It includes a multi-model switcher offering both OpenAI and Anthropic models. However, independent data documents a gap between integration and performance: Microsoft's own published figure for its Excel agent on SpreadsheetBench is 57.2%, and in the published 11-task speed head-to-head it never finished the 10,000-row job, averaging nearly 2 minutes per task versus roughly 16 seconds for the fastest competitor. Pricing runs $18-32 per user for business tiers and $19.99 for consumer plans. For governed Microsoft 365 environments where native integration and compliance matter more than raw speed, it's still the path of least resistance in 2026.

Best for: Governed Microsoft 365 environments that want native, compliant AI inside Excel with zero extra tooling.

Pricing: $18-32/user/month business tiers; $19.99/month consumer

Key features

• Agentic AI built directly into Excel
• Works on local Excel files as of the 2026 desktop update
• Multi-model switcher with OpenAI and Anthropic models
• Generates formulas, analysis and charts from a prompt
• No separate install or add-in required

Pros

• Deepest possible integration — it lives inside Excel
• Local file support since the 2026 update
• Multi-model flexibility via the model picker
• Fits existing Microsoft 365 licensing and governance

Cons

• 57.2% SpreadsheetBench accuracy per Microsoft's own published figure
• Never finished the 10K-row job in the published speed head-to-head
• Slower per task than leading competitors (nearly 2 minutes vs ~16 seconds)
• Requires Microsoft 365 paid tiers

Visit Edit with Copilot (Microsoft Excel)

#7 Open-Source Spreadsheet Skills (ai-config-table-skill, TabClaw, CSV Data Summarizer)

The open-source community has produced a wave of genuinely useful spreadsheet skills for 2026. The ai-config-table-skill lets AI agents safely edit Excel, CSV, TSV and JSON config tables with a scan → formula gate → preview → copy → diff → user-confirmation workflow that never overwrites the source until you confirm. TabClaw, from USTC's table-mining group, is an interactive, self-evolving agent for spreadsheet manipulation and table reasoning — drop in a CSV or Excel file, describe what you want, and it supports interactive planning, parallel multi-table reasoning, persistent memory and reusable skill distillation. The csv-data-summarizer-claude-skill automatically analyzes CSV files with summary statistics, missing-data detection and visualizations like time-series plots, correlation heatmaps and distribution histograms, adapting to domains from e-commerce to finance. These skills are free, inspectable and portable across SKILL.md-compatible agents — but that same openness is exactly why independent verification and security scanning matter before you equip them.

Best for: Developers and technical teams comfortable inspecting and installing skills directly from GitHub.

Pricing: Free and open source

Key features

• ai-config-table-skill: safe config-table editing with formula gate, dry-run patch, diff and user-confirmation workflow
• Never overwrites source data until the user confirms
• TabClaw: interactive spreadsheet manipulation with parallel multi-table reasoning and persistent memory
• TabClaw supports reusable skill distillation from your workflows
• CSV Data Summarizer: summary statistics, missing-data detection and visualizations (time-series, heatmaps, histograms)
• Free, open-source and portable across SKILL.md-compatible agents

Pros

• Completely free and fully inspectable source code
• Strong safety patterns like confirmation-gated edits and diffs
• Portable across Claude Code, Codex, Cursor, Gemini CLI and other SKILL.md-compatible agents
• Domain-flexible — from config tables to e-commerce and finance data

Cons

• No independent benchmarking — you can't verify performance claims before installing
• Open-source skills are a documented supply-chain risk without security scanning
• Requires technical comfort to evaluate, install and maintain
• Maintenance depends on volunteer authors; upstream changes can break workflows silently

Visit Open-Source Spreadsheet Skills (ai-config-table-skill, TabClaw, CSV Data Summarizer)

Conclusion

The spreadsheet agent market in 2026 has matured past hype into measurable territory: independent benchmarks like SpreadsheetBench now separate tools that complete 92.5% of real-world tasks from those that don't finish large jobs at all, and the agent-skills ecosystem has grown to over a million listings across registries. But more choice has made the selection problem harder, not easier — and with published research finding roughly 26% of skills carry at least one vulnerability, equipping your agent with the wrong spreadsheet skill isn't just inefficient, it's a security decision. That's why SkillzBench leads this list: it's being built to independently discover, verify, benchmark and equip agent skills, publishing Lift in percentage points and Trust scores that money can't buy, so you know a spreadsheet skill actually helps before it touches your workbooks. If you're ready to stop guessing and start proving which skills earn their place in your agents' loadouts, join the SkillzBench waitlist today. Skills, proven.

Frequently asked questions

What makes SkillzBench different from other agent skill platforms?

SkillzBench is an independent testing lab and control panel, not a skill marketplace. It's being built around five pillars — Discover, Verify, Benchmark, Equip and Stay current — that together let you see a skill's security posture, license status, maintenance health and measured Lift (published in percentage points from sandboxed with-and-without testing) before you equip it. Its Trust score (0-100) draws from security, quality, license, maintenance and community signals, Lift is always shown separately from ratings, and money never touches scores. Note that SkillzBench is pre-launch: the waitlist and newsletter are live, and benchmarking, scanning and the MCP server describe what the platform will offer at launch.

What are agent skills for Excel and spreadsheets?

Agent skills are instruction packages — typically SKILL.md files with reference material — that teach AI agents like Claude Code, Codex, Cursor, Gemini CLI and chat-app agents how to perform specialized tasks. Spreadsheet-focused skills cover things like XLSX/CSV manipulation, formula generation, dataset quality auditing, config-table editing and data visualization. The agent reads the skill's short description, decides when it's relevant, and loads the full instructions on demand, keeping your context window clean.

Are open-source spreadsheet skills safe to install?

They can be, but caution is warranted. Published research found roughly 26% of skills carry at least one vulnerability, and community discussion in 2026 has flagged unvetted SKILL.md files as a supply-chain risk. Some open-source skills, like ai-config-table-skill, are designed with safety patterns such as confirmation-gated edits and diffs, but no independent benchmark or security scan exists for most. That's the gap SkillzBench is being built to close: commit-SHA pinning, security scanning, license checks and sandboxed benchmarking before you equip anything.

Which AI agent is most accurate for Excel in 2026?

According to independent data, GPT for Excel (GPT for Work) posted the strongest documented commercial score — 92.5% on SpreadsheetBench V1-Verified and 97.5% in AIMultiple's test. Claude for Excel scored 95% in AIMultiple's testing and leads on complex model audits and formula tracing. Microsoft's own published figure for its Excel agent is 57.2% on SpreadsheetBench. Accuracy varies significantly by task, so match the tool to your specific workflow rather than relying on a single ranking.

How much do spreadsheet AI agents and skills cost in 2026?

Pricing varies widely. Open-source skills are free. GPT for Excel uses pay-per-use from $29 with no subscription. Claude for Excel is included in Claude Pro at $20/month. Edit with Copilot runs $18-32 per user per month for business tiers and $19.99 for consumer plans. SkillzBench's planned tiers include a generous free tier, Pro (~$9-15/mo), Teams (~$15-25/seat/mo) and Creator (~$5-10/mo). For tools without published pricing, contact the vendor directly.

Can I use the same agent skill across different AI agents?

Yes — that's one of the biggest advantages of the SKILL.md format. The same skill files generally work across Claude Code, Codex CLI, Cursor, Gemini CLI and other compatible platforms, with install paths varying per agent. SkillzBench's planned Equip pillar takes this further with one MCP server plus per-agent loadouts: you toggle which skills and tools each agent can see, so each agent only gets the capabilities you've chosen for it.